Lync Front End and VMware CPU Ready %

Question

We are running an Enterprise Lync 2013 environment with Voice. It's running in on ESX with 3 front end boxes and 2 edge servers. Every now and then we get users complaining of poor audio quality on conference calls. Although there can be many reasons for
this that are not "Technology" related we have been digging into it. We noticed that our CPU ready % was kind of high on the Front End boxes. It was pushing 10%... after some ongoing back and forth with our VMWare engineers we got them to isolate
our boxes for a few weeks to see if that helped. It seemed to have done the trick but our VMware engineers don't like wasting resources so they put Lync back in with general population. Now CPU ready % is hovering more around 5% so it's better than it was
before. My questions is what % CPU ready are you guys calling "Too High" in your environments?

Answer 1

Thanks for your response Ben. We have gone through the white paper with our ESX engineer. Unfortunately they don't think it applies to ESX because it focuses on HyperV. Although much of it is applicable. My hope in this post is just to get a community benchmark
on what people are finding in their environments that "work well" for them. The main topic we have been pushing is this part of the Virtualization White Paper.
8.1.11 Resource Over-Allocation
Lync Server 2013 guests should not be provisioned on hypervisor hosts which are configured for CPU over allocation. For example, if the physical host has 24 cores, no more than 24 cores total should
be allocated to all of the guests on the system.
Lync Server includes several real-time workloads (such as audio/video and conferencing) that require real-time access to components such as processor, memory, network, and storage. If these components
are shared among other guests and Lync does not have access to these as required, the result can be a negative user experience including dropped calls, dropped audio, choppy audio, inability to join a conference, paused video, and other user-noticeable results.
These can be difficult to troubleshoot, as an analysis of the physical servers and virtual guests at a later point in time may show all is fine, with the issues surfacing only during periods of high utilization from other guests.